The Cognitive Mechanism and Limitations of Deep Learning from the Perspective of the Three-Level Cognitive Theory
- 1 School of Philosophy, Beijing Normal University, Beijing, China
Abstract
Building on the Three-Level Theory of Cognition, this paper examines the architecture and foundational principles of deep learning in order to clarify its specific cognitive mechanisms and the central epistemological challenges it poses. The cognitive efficacy of deep learning derives from the synergistic interaction of three distinct levels: computational theory defines its objectives, algorithms specify the methods for achieving them, and hardware serves as the physical substrate for algorithmic implementation. Together, these levels jointly drive the advancement of deep learning. Meanwhile, the current challenges faced by deep learning, namely the lack of interpretability and theoretical understanding, also constitute the core epistemological issues in machine learning. By identifying these constraints at each hierarchical level, this paper highlights where future research must focus to deepen our understanding of deep learning’s cognitive capabilities and to resolve its foundational epistemic problems.
- Biederman, I. (1987). Recognition-by-Components: A Theory of Human Image Understanding. Psychological Review, 94, 115-147. https://doi.org/10.1037/0033-295x.94.2.115
- Budach, L., Feuerpfeil, M., Ihde, N., Nathansen, A., Noack, N. S., Patzlaff, H., Harmouch, H., & Naumann, F. (2022). The Effects of Data Quality on ML-Model Performance . https://arxiv.org/abs/2207.14529
- Fodor, J. A., & Pylyshyn, Z. W. (1988). Connectionism and Cognitive Architecture: A Critical Analysis. Cognition, 28, 3-71. https://doi.org/10.1016/0010-0277(88)90031-5
- Gao, Y., Dong, K., Shan, C., Li, D., & Liu, Q. (2025). Causal Disentanglement for Single-Cell Representations and Controllable Counterfactual Generation. Nature Communications, 16, Article No. 6775. https://doi.org/10.1038/s41467-025-62008-1
- Goodfellow, I., Bengio, Y., & Courville, A. (2016). Deep Learning . MIT Press. https://www.deeplearningbook.org/lecture_slides.html
- Li, X., Rosman, G., Gilitschenski, I., DeCastro, J. A., Vasile, C. I., Karaman, S., & Rus, D. (2020). Differentiable Logic Layer for Rule Guided Trajectory Prediction. In 4th C onference on Robot Learning . https://tisl.cs.toronto.edu/publication/202011-corl-logic_layer/corl20-logic_layer.pdf
- Lipton, Z. C. (2016). The Mythos of Model Interpretability. Communications of the ACM, 61, 36-43. https://doi.org/10.1145/3233231
- Marr, D. (1982). Vision: A Computational Investigation into the Human Representation and Processing of Visual Information . MIT Press.
- Pylyshyn, Z. W. (1986). Computation and Cognition: Toward a Foundation for Cognitive Science . The MIT Press.
- Simon, H. A., & Newell, A. (1971). Human Problem Solving: The State of the Theory in 1970. American Psychologist, 26, 145-159. https://doi.org/10.1037/h0030806
- von Eschenbach, W. J. (2021). Transparency and the Black Box Problem: Why We Do Not Trust AI. Philosophy & Technology, 34, 1607-1622. https://doi.org/10.1007/s13347-021-00477-0
- Xu, S., Ge, Y., & Zhang, Y. (2023). Causal Explainable AI. In S. Li, & Z. Chu (Eds.), Machine Learning for Causal Inference (pp. 137-159). Springer International Publishing. https://doi.org/10.1007/978-3-031-35051-1_7